INTERFERED RUSSIAN SPEECH OF NATIVE SPEAKERS OF THE KABARDIAN-CIRCASSIAN LANGUAGE: AN EXPERIMENTAL STUDY

Abstract

The mismatch between the speech of non-native’s and the native language resources used in training, as well as acoustic and linguistic modelling of pronunciation based on a data-driven approach induce the challenging task of automatic speech recognition for non-native’s. The existing approaches to overcoming these difficulties are based on the use of non-native speech resources or multilingual corpora and are proved to be ineffective. It is necessary to study the specifics of speech psychophysiological mechanisms of bilinguals who use two or more language systems in communication. This paper shows the results of the experimental phonetic study of interlingual interference and accent in Russian speech of the native Kabardian-Circassian speaker s. The experimental materials include audio recordings of phonatory reading of a previously prepared list of words. The preliminary acoustic analysis of the initial formant measurements executed on the non-native Russian speech of the Kabardian-Circassian speakers showed a systematic deviation of the acoustic patterns of [A] and [O] vowel allophones in the F1-F2 space. The crucial identification feature of the accent vowel [A] in the speech of male and female native speakers of the Kabardian-Circassian language is the systematic deviation of its formant characteristics in space (F1-F2) to the right-up direction from the standard values of the orthoepic norm of the Russian language. The identification features of vowel [O] in the speech of male speakers are the high variability of the F2 second formant values and a significant shift of the average value of F2 upwards from the gender-neutral and gender-specific reference values. For female speakers, the distinctive feature of the Kabardian-Circassian accent is the localisation of observations to the right by F1 and down by F2 relative to the gender-specific reference in space (F1-F2). Comparison of vocal systems based on acoustic characteristics allows the authors to assess objectively the mutual influence of language systems, identify systematic errors in real speech acti vity, and reduce the inaccuracy of categorisation based on rather conventional phonetic features of vowels. This research is aimed at developing methods for modelling the speech of non-native speakers for subsequent use in speech recognition systems, as well as accent and language identification in conditions of limited linguistic resources. The experimental data obtained can be useful for further development of contrastive analysis theory, typical interference pronunciation errors identification, and also contribute to elaboration of theoretical models of foreign language acquisition.